Research And Implementation Of The Personalized Recommendation System Based On Tags

With the rapid development of the electronic commerce and Internet, more and more consumers lean how to browse information resources on the Internet which do great help to them. However, they often feel confused in the various resource space with the multiple choice and can not quickly find their own needs.With the help of the personalized recommendation system, this problem has been greatly solved.The collaborative filtering recommendation algorithm is the mainstream recommendation algorithm, but with the increasing number of users and goods and expanding scale of the website, many serious problems caused the decline in the quality of recommendation system, such as sparsity problem and Cold-Start problem and so on.This paper presents an improved collaborative filtering recommendation algorithm based on Preference Tag named CFBPT algorithm. Firstly extract of emotion labels from the user comments data, then combine the preference tags with score data to produce recommendation results to alleviate the sparsity problem and user cold start problem of the traditional collaborative filtering recommendation algorithm. And at last the experimental results show that the results of this study have the effectiveness and feasibility.The main work of this paper is as follows:(1) Extract preference tags from the user reviews. The traditional definition of tag refers to the keywords which used to describe information and generated by professional researchers or some users. This paper presents a new tag which comes from the user comments and then regarded as the data to improve the CF algorithm.(2) For the item neighbor set is not accurate, this paper proposes an improved comprehensive similarity. This paper will put review mining technology into the collaborative filtering algorithm. Based on the Apriori algorithm and the syntactic template technology, extract feature words and emotional words to form a new matrix item-feature matrix, and then combine the common user rating to calculate the item similarity. Experiment shows that the neighbor set is more accurate than before.(3) In order to solve the sparse problem, fill the prediction score into the user-item matrix, and then get the recommend result. The experiment results show that the improved CFBPT algorithm improves the accuracy of the recommend results.(4) In order to solve the cold-start problem, construct the item-feature vector of the new user to look for the similar cluster.(5) Finally build a personalized recommendation system and do some experiments of the improved algorithm to get the conclusion.